feat(tui): live GPU chart during stress test, full VRAM allocation

- GPU Platform Stress Test now shows a live in-TUI chart instead of nvtop.
  nvidia-smi is polled every second; up to 60 data points per GPU kept.
  All three metrics (Usage %, Temp °C, Power W) drawn on a single plot,
  each normalised to its own range and rendered in a different colour.
- Memory allocation changed from MemoryMB/16 to MemoryMB-512 (full VRAM
  minus 512 MB driver overhead) so bee-gpu-stress actually stresses memory.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

audit/internal/tui/update.go

@@ -3,6 +3,7 @@ package tui
 import (
 	"fmt"
 	"strings"
+	"time"
 
 	tea "github.com/charmbracelet/bubbletea"
 )
@@ -130,6 +131,22 @@ func (m model) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
 			m.body = msg.body
 		}
 		return m, m.refreshSnapshotCmd()
+	case gpuLiveTickMsg:
+		if m.screen == screenGPUStressRunning {
+			if len(msg.rows) > 0 {
+				elapsed := time.Since(m.gpuLiveStart).Seconds()
+				for i := range msg.rows {
+					msg.rows[i].ElapsedSec = elapsed
+				}
+				m.gpuLiveRows = append(m.gpuLiveRows, msg.rows...)
+				n := max(1, len(msg.indices))
+				if len(m.gpuLiveRows) > 60*n {
+					m.gpuLiveRows = m.gpuLiveRows[len(m.gpuLiveRows)-60*n:]
+				}
+			}
+			return m, pollGPULive(msg.indices)
+		}
+		return m, nil
 	case nvidiaSATDoneMsg:
 		if m.nvidiaSATAborted {
 			return m, nil